ProtoInfoMax: Prototypical Networks with Mutual Information Maximization for Out-of-Domain Detection

TL;DR

ProtoInfoMax enhances out-of-domain detection in NLP by extending Prototypical Networks with a mutual information maximization objective, significantly improving performance and reliability in low-resource scenarios.

Contribution

It introduces ProtoInfoMax, a novel architecture that combines prototypical networks with mutual information maximization for better OOD detection.

Findings

Improves OOD detection accuracy by up to 20% in low-resource settings.

Reduces neural network overconfidence errors.

Enhances reliability of predictions in NLP applications.

Abstract

The ability to detect Out-of-Domain (OOD) inputs has been a critical requirement in many real-world NLP applications. For example, intent classification in dialogue systems. The reason is that the inclusion of unsupported OOD inputs may lead to catastrophic failure of systems. However, it remains an empirical question whether current methods can tackle such problems reliably in a realistic scenario where zero OOD training data is available. In this study, we propose ProtoInfoMax, a new architecture that extends Prototypical Networks to simultaneously process in-domain and OOD sentences via Mutual Information Maximization (InfoMax) objective. Experimental results show that our proposed method can substantially improve performance up to 20% for OOD detection in low resource settings of text classification. We also show that ProtoInfoMax is less prone to typical overconfidence errors of Neural Networks, leading to more reliable prediction results.